Github user mengxr commented on a diff in the pull request:
https://github.com/apache/spark/pull/9848#discussion_r45527879
--- Diff:
mllib/src/main/scala/org/apache/spark/ml/tuning/CrossValidator.scala ---
@@ -131,6 +137,181 @@ class CrossValidator(override val uid: String)
extends Estimator[CrossValidatorM
}
copied
}
+
+ // Currently, this only works if all [[Param]]s in
[[estimatorParamMaps]] are simple types.
+ // E.g., this may fail if a [[Param]] is an instance of an [[Estimator]].
+ // However, this case should be unusual.
+ @Since("1.6.0")
+ override def write: MLWriter = new
CrossValidator.CrossValidatorWriter(this)
+}
+
+@Since("1.6.0")
+object CrossValidator extends MLReadable[CrossValidator] {
+
+ @Since("1.6.0")
+ override def read: MLReader[CrossValidator] = new CrossValidatorReader
+
+ @Since("1.6.0")
+ override def load(path: String): CrossValidator = super.load(path)
+
+ private[CrossValidator] class CrossValidatorWriter(instance:
CrossValidator) extends MLWriter {
+
+ SharedReadWrite.validateParams(instance)
+
+ override protected def saveImpl(path: String): Unit =
+ SharedReadWrite.saveImpl(path, instance, sc, JNothing)
+ }
+
+ private class CrossValidatorReader extends MLReader[CrossValidator] {
+
+ /** Checked against metadata when loading model */
+ private val className = classOf[CrossValidator].getName
+
+ override def load(path: String): CrossValidator = {
+ val (metadata, estimator, evaluator, estimatorParamMaps, numFolds) =
+ SharedReadWrite.load(path, sc, className)
+ new CrossValidator(metadata.uid)
+ .setEstimator(estimator)
+ .setEvaluator(evaluator)
+ .setEstimatorParamMaps(estimatorParamMaps)
+ .setNumFolds(numFolds)
+ }
+ }
+
+ private object CrossValidatorReader {
+ /**
+ * Examine the given estimator (which may be a compound estimator) and
extract a mapping
+ * from UIDs to corresponding [[Params]] instances.
+ */
+ def getUidMap(instance: Params): Map[String, Params] = {
+ val subStages: Array[Params] = instance match {
+ case p: Pipeline => p.getStages.asInstanceOf[Array[Params]]
+ case pm: PipelineModel => pm.stages.asInstanceOf[Array[Params]]
+ case v: ValidatorParams => Array(v.getEstimator, v.getEvaluator)
+ case _: Params => Array()
+ }
+ val subStageMaps =
subStages.map(getUidMap).foldLeft(Map.empty[String, Params])(_ ++ _)
--- End diff --
* minor: Use `List` in recursive calls might be cheaper.
* If we use Map, shall we detect duplicate IDs and throw an warning? It is
likely to be wrong if there are two stages with the same ID in a pipeline.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
